U.S. patent application number 12/806316 was filed with the patent office on 2011-03-10 for method for self-adjustment of a triaxial acceleration sensor during operation, and sensor system having a three -dimentional acceleration sensor.
Invention is credited to Alexander Buhmann, Axel Franke.
Application Number | 20110060543 12/806316 |
Document ID | / |
Family ID | 43535904 |
Filed Date | 2011-03-10 |
United States Patent
Application |
20110060543 |
Kind Code |
A1 |
Franke; Axel ; et
al. |
March 10, 2011 |
Method for self-adjustment of a triaxial acceleration sensor during
operation, and sensor system having a three -dimentional
acceleration sensor
Abstract
A method for self-adjustment of a triaxial acceleration sensor
during operation includes: calibrating the sensor; checking the
self-adjustment for an interfering acceleration, with the aid of a
measurement equation and estimated values for sensitivity and
offset; repeating the adjustment if an interfering acceleration is
recognized; and accepting the estimated values for sensitivity and
offset as calibration values if an interfering acceleration is not
recognized. The step of checking the self-adjustment includes:
estimating sensitivity and/or offset and the variance thereof;
determining an innovation as the difference between a measured
value of the measurement equation and an estimated value of the
measurement equation; testing the innovation for a normal
distribution; and recognizing the interfering acceleration in the
event of a deviation from the normal distribution.
Inventors: |
Franke; Axel; (Ditzingen,
DE) ; Buhmann; Alexander; (Stuttgart, DE) |
Family ID: |
43535904 |
Appl. No.: |
12/806316 |
Filed: |
August 9, 2010 |
Current U.S.
Class: |
702/104 |
Current CPC
Class: |
G01P 21/00 20130101 |
Class at
Publication: |
702/104 |
International
Class: |
G01P 21/00 20060101
G01P021/00 |
Foreign Application Data
Date |
Code |
Application Number |
Sep 4, 2009 |
DE |
10 2009 029 216.0 |
Claims
1. A method for self-adjustment of a triaxial acceleration sensor
during operation, comprising: calibrating the sensor for
sensitivity and offset with the aid of calibration values; checking
for an interfering acceleration, with the aid of a measurement
equation and estimated values for sensitivity and offset, wherein
the checking includes: estimating sensitivity, variance of
sensitivity, offset and variance of offset; determining a
difference between a measured value of the measurement equation and
an estimated value of the measurement equation; testing the
determined difference for a normal distribution; and recognizing
that an interfering acceleration exists in the event of a deviation
from the normal distribution; repeating the calibrating step if an
interfering acceleration is recognized; and accepting the estimated
values for sensitivity and offset as calibration values if an
interfering acceleration is not recognized.
2. The method as recited in claim 1, further comprising:
calculating and outputting an acceleration value.
3. The method as recited in claim 2, wherein the sensitivity,
variance of sensitivity, offset and variance of offset are
estimated using an estimator.
4. The method as recited in claim 3, wherein a Kalman filter is
used for estimating the sensitivity, variance of sensitivity,
offset and variance of offset.
5. The method as recited in claim 3, wherein the measurement
equation describes the absolute value of the acceleration as
corresponding to 1 g.
6. The method as recited in claim 5, wherein in determining the
difference between a measured value of the measurement equation and
an estimated value of the measurement equation, the absolute value
of an acceleration vector is estimated to be equal to 1 g, and a
measured value of 1 g is assumed as the acceleration.
7. The method as recited in claim 5, wherein in determining the
difference between a measured value of the measurement equation and
an estimated value of the measurement equation, a normalized
difference is used, and wherein in testing the determined
difference for a normal distribution, the normalized difference is
tested for a chi square distribution.
8. The method as recited in claim 5, wherein before the step of
calibrating the sensor for sensitivity and offset, the method
further includes ensuring the observability of the sensitivity and
the offset of the sensor by recognizing a rest situation.
9. The method as recited in claim 8, wherein the step of ensuring
the observability of the sensitivity and the offset of the sensor
further includes recognizing whether new information is
present.
10. A sensor system, comprising: a three-dimensional acceleration
sensor; a memory; an estimator; and a computing unit configured to
perform a calibration of the acceleration sensor during operation
of the sensor, wherein the computing unit is configured to (a)
perform the calibration with the aid of the estimator, and (b) test
a distribution function.
11. The sensor system as recited in claim 10, wherein values of a
zero error and a sensitivity of the acceleration sensor are stored
in the memory.
12. The sensor system as recited in claim 11, wherein the values of
the zero error and the sensitivity are stored in the memory during
manufacture of the sensor system.
13. The sensor system as recited in claim 11, further comprising an
ASIC.
14. The sensor system as recited in claim 11, wherein the sensor
system is configured as a module.
15. The sensor system as recited in claim 11, wherein the computing
unit is an external computing unit.
16. The sensor system as recited in claim 11, wherein the computing
unit is configured to perform the calibration in real time during
operation of the sensor.
Description
BACKGROUND OF THE INVENTION
[0001] 1. Field of the Invention
[0002] The present invention is directed to a method for
self-adjustment of a triaxial acceleration sensor and a sensor
system having a three-dimensional acceleration sensor during
operation.
[0003] 2. Description of Related Art
[0004] Micromechanical acceleration sensors are known, and are
widely used in particular as acceleration or rotational rate
sensors. The sensors must be adjusted to their field of
application, using the higher-order process control system. The
adjustment is usually carried out with a certain level of effort at
the end of the manufacturing process by accelerating the sensor in
its sensitive spatial axis, for all sensor axes in succession.
These types of sensors have the disadvantage that a drift from the
zero point and sensitivity during operation are not taken into
account. Another approach is the use of the gravitational vector as
a reference for an adjustment during operation, described in
Loetters et al.: "Procedure for in-use calibration of triaxial
accelerometers in medical applications," Sensors and Actuators A 68
(1998), 221-228. A method therein is based on the fundamental
principle that the sensor on a patient is not constantly
accelerated, but instead has rest phases in which the acceleration
of gravity may be used for the calibration. The method essentially
uses the following method steps, not necessarily in the order
stated: [0005] ensuring the observability of the sensitivity and
the offset; [0006] calibrating the sensor with the aid of
calibration values for sensitivity and offset; [0007] checking the
self-adjustment for an interfering acceleration, with the aid of a
measurement equation and estimated values for sensitivity and
offset; [0008] repeating the adjustment if an interfering
acceleration is recognized; [0009] accepting the estimated values
for sensitivity and offset as calibration values if an interfering
acceleration is not recognized.
[0010] Ensuring the observability of the sensitivity and the offset
of the sensor means the recognition of a time interval which may be
a candidate for a rest phase, so that sensitivity and offset may be
determined as calibration values from the measured data. According
to Loetters et al., the use of various filters ensures the
observability and recognition of an interfering acceleration. One
disadvantage of this method is that the filters in particular must
be adapted to the product scenario, for example with respect to
cut-off frequencies. This requires additional modeling, and limits
the sensor to the particular scenarios. Loetters et al. use a
sensor system having a three-dimensional acceleration sensor, a
computing unit, and a memory, the computing unit being designed to
carry out a calibration of the acceleration sensor during
operation.
BRIEF SUMMARY OF THE INVENTION
[0011] In contrast, the method according to the present invention
and the sensor system according to the present invention have the
advantage that an observer recognizes interfering accelerations
with the aid of statistical tests which do not have to be adapted
to the product scenario. Sensor errors are estimated and corrected
using estimators, for example Kalman filters and error square
minimization algorithms. This results in the additional advantages
that adjustment is not necessary at the end of the sensor
manufacturing process, and that influences on the sensor
parameters, such as temperature fluctuations and aging of the
sensor, are implicitly taken into account. The present invention
allows stipulation of stricter specifications, which may be
maintained under external influences. A further advantage of the
present invention is the reduction of testing costs.
[0012] A method according to the present invention for
self-adjustment of a triaxial acceleration sensor during operation,
having the following method steps: [0013] b. calibrating the sensor
with the aid of calibration values for sensitivity and
offset--sensitivity and offset are determined; [0014] c. checking
the self-adjustment for an interfering acceleration with the aid of
a measurement equation and estimated values for sensitivity and
offset--a check is made as to whether an interfering acceleration
has occurred during a calibration measurement; [0015] d. repeating
the adjustment if an interfering acceleration is recognized--in the
event of an interfering acceleration the measured values are not
suitable for a calibration; [0016] e. accepting the estimated
values for sensitivity and offset as calibration values if an
interfering acceleration is not recognized--if no interfering
acceleration is present, the measured values are suitable for a
calibration, and the estimated values for sensitivity and offset
are used as calibration data; [0017] has the advantage according to
the present invention that method step c. has the following
substeps: [0018] c1. estimating sensitivity and/or offset and the
variance thereof; [0019] c2. determining an innovation as the
difference between a measured value of the measurement equation and
an estimated value of the measurement equation; [0020] c3. testing
the innovation for a normal distribution; [0021] c4. recognizing
the interfering acceleration in the event of a deviation from the
normal distribution.
[0022] In the subdivision of the procedure into method steps, it is
pointed out that the individual steps are tuned to one another, and
that different interactions may occur, depending on the embodiment.
Thus, an interfering acceleration is any acceleration which
deviates from a static state, i.e., rest or constant velocity, and
thus practically any acceleration which differs from gravitational
acceleration. This is frequently the acceleration which the sensor
is intended to measure in the particular application, and which in
such a case is output in method step d. in a corresponding
advantageous embodiment. However, such an acceleration is an
interference for the calibration. On the one hand, a rest phase of
the sensor is defined as the absence of an interfering
acceleration, which in one embodiment described below is relevant
for observability in step a., and on the other hand, an interfering
acceleration is recognized or its absence is verified in step c.,
in particular step c4. Substeps of method step c. in combination
characterize the method according to the present invention, but
with regard to the overall method may be used under one or several
other method steps, so that their combination according to the
present invention is still carried out in cooperation.
[0023] In one advantageous embodiment of the present invention, an
estimator is used for estimating sensitivity and/or offset and the
variance thereof. A Kalman filter is preferably used for estimating
sensitivity and/or offset and the variance thereof. In this case
the extended Kalman filter (EKF), as an iterative filter, is
recommended as advantageous for real-time applications, and the
unscented Kalman filter (UKF) is recommended for use with highly
nonlinear functions.
[0024] The measurement equation advantageously describes the
absolute value of the acceleration as corresponding to 1 g. In
method step c2. the absolute value of an acceleration vector is
preferably estimated to be equal to 1 g, and a measured value of 1
g is assumed for the acceleration.
[0025] In another advantageous embodiment of the present invention,
a normalized innovation is used in method step c2., and in method
step c3. the normalized innovation is tested for a chi square
distribution instead of the normal distribution.
[0026] Another advantageous embodiment of the method according to
the present invention includes, before method step b., method step
a., ensuring the observability of the sensitivity and offset of the
sensor, having substep a1, recognizing a rest situation. Method
step a. preferably also includes substep a2., recognizing whether
new information is present. Using this method step, a situation may
be recognized in which a sensor alternates between only two
positions over a fairly long period of time. These two positions do
not provide information, since their gravitational vectors point in
directions which have already been taken into account. The
calibration of the sensor may drift without a false positive being
recognized. According to the present invention, recognizing the new
measured values as information which is not new may result in
discarding of the new measured values for the statistics in order
to avoid a false positive. Here as well, a statistical test may
advantageously be used to ensure the observability. An observer
estimates the state of the measuring device on the basis of one
measurement, and the sensitivity and offset are determined from the
measured data.
[0027] New information may be recognized using two approaches. The
first approach is based on an analysis of the measured data or the
estimated acceleration in a Cartesian coordinate system. A null
hypothesis is established that two states originate from the same
distribution function having a normal distribution. The null
hypothesis may be proved or disproved with the aid of a z test and
a suitable test variable. The second approach is based on the
transformation of the measured data into polar coordinates. This
allows an intuitive consideration of the new measured value. This
approach is disadvantageous if zero errors which are present cause
the radius to become small in polar coordinates, since in that case
any data which are not new might be recognized as new. However,
this approach offers advantages in conjunction with a Kalman
filter, which estimates the gravitational acceleration.
[0028] A sensor system according to the present invention, having a
three-dimensional acceleration sensor, a computing unit, and a
memory, the computing unit being designed to carry out a
calibration of the acceleration sensor during operation, has the
advantage that the computing unit is designed to carry out the
calibration of the acceleration sensor with the aid of an
estimator, and to test a distribution function. As the result of
using statistical methods there is no need to use
application-specific filters.
[0029] In one example embodiment of the present invention, values
of a zero error and/or of a sensitivity of the acceleration sensor
are stored in the memory of the sensor system. These values may be
used as starting values for a calibration during operation. The
values of the zero error and/or of the sensitivity are
advantageously stored in the memory during manufacture of the
sensor system.
[0030] In one advantageous embodiment of the present invention the
sensor system has an ASIC. This makes a compact sensor system
having a sensor and an integrated evaluation unit possible.
[0031] In one alternative advantageous embodiment of the present
invention, the sensor system has an external computing unit.
[0032] In this case, a computing unit which is already present for
other purposes may perform the calibration of the sensor during
operation.
[0033] The calibration preferably occurs in real time.
BRIEF DESCRIPTION OF THE DRAWINGS
[0034] FIG. 1 shows a schematic representation of a sensor system
according to the present invention, in a first example embodiment
having an external computing unit.
[0035] FIG. 2 shows a schematic representation of a sensor system
according to the present invention, in a second example embodiment
having an integrated computing unit.
[0036] FIG. 3 shows a flow chart of the method according to the
present invention in a first. example embodiment.
[0037] FIG. 4 shows a flow chart of the method according to the
present invention in a second example embodiment.
DETAILED DESCRIPTION OF THE INVENTION
[0038] FIG. 1 shows a sensor system 20 according to the present
invention, having a three-dimensional acceleration sensor 21, and
an external evaluation unit 22 having a computing unit 23 and a
memory 24. Computing unit 23 has an estimator 25 which is
advantageously implemented as software. Computing unit 23 is able
to carry out a calibration of acceleration sensor 21 with the aid
of estimator 25, and to carry out testing of a distribution
function in real time. Acceleration sensor 21 is represented in
three dimensions in its Cartesian coordinate system via axes x, y,
z in the measuring directions of the sensor. In a rest state,
gravity acts on the sensor in the direction of arrow 26, with the
absolute value of 1 g of the gravitational acceleration.
[0039] FIG. 2 shows a sensor system 30 according to the present
invention in one specific embodiment as an independent integrated
module 31 having a three-dimensional acceleration sensor 32, and an
evaluation unit 33, in the form of an ASIC 37, having a computing
unit 34 and a memory 35. Computing unit 34 is designed to carry out
a calibration of acceleration sensor 32 with the aid of an
estimator 36, and to test a distribution function in real time.
Values of a zero error and of a sensitivity of the acceleration
sensor ascertained during manufacture of the sensor system are
stored in memory 35.
[0040] FIG. 3 shows in flow chart 40 a method for self-adjustment
of a triaxial acceleration sensor during operation, having the
following method steps: [0041] b. calibrating the sensor with the
aid of calibration values for sensitivity and offset; [0042] c.
checking the self-adjustment for an interfering acceleration with
the aid of a measurement equation and estimated values for
sensitivity and offset, having the following substeps: [0043] c1.
estimating sensitivity and/or offset and the variance thereof--a
Kalman filter is employed, and is used as an estimator for
estimating sensitivity and offset and the variance thereof; [0044]
c2. determining an innovation as the difference between a measured
value of the measurement equation and an estimated value of the
measurement equation--the measurement equation describes the
absolute value of the acceleration as corresponding to 1 g; [0045]
c3. testing the innovation for a normal distribution; [0046] c4.
recognizing the interfering acceleration in the event of a
deviation from the normal distribution, in addition to the
following method step: [0047] d. repeating the adjustment if an
interfering acceleration is recognized, as indicated by arrow 41;
[0048] e. accepting the estimated values for sensitivity and offset
as calibration values if an interfering acceleration is not
recognized. The method is repeated during operation, as indicated
by arrow 42. In the next pass, the values for sensitivity and
offset which have just been accepted in step e. are used in step b.
as calibration values. The calibration is thus improved according
to the present invention, and in the event of a temperature drift
or an aging drift of the sensor the calibration is updated.
[0049] Regarding the subdivision of the procedure into method
steps, it is pointed out that the individual steps are tuned to one
another, and that different interactions may occur, depending on
the embodiment. Substeps of method step c. in combination
characterize the method according to the present invention, but
with regard to the overall method may be used under one or several
other method steps, so that their combination according to the
present invention is still carried out in cooperation. Thus,
depending on whether an interfering acceleration has occurred
during or after the calibration, step c. may be carried out during
or after step b.; i.e., substeps of method step c. may be concluded
before method step b. is completed.
[0050] In the Cartesian coordinate system of acceleration sensor 21
from FIG. 1, acceleration a of the rest states in all positions of
the sensor in space is situated on a spherical surface having a
radius of 1 g. If a switch is made to output voltage U supplied by
the sensor, the sphere is elliptically deformed according to the
different sensitivities of the measuring directions as described in
sensitivity matrix S. The zero errors result in an offset O, which
may be represented as a displacement of the ellipsoid from the
origin of the coordinate system. In the system which is assumed to
be linear, the expression U=S*a+O applies for the output voltage,
resulting in a=S.sup.-1 (U-O) for the acceleration acting on a test
mass. For simplification, S is assumed to be a diagonal matrix. For
a rest state, also referred to as a quasi-static state, the
following expression is valid: |a|=1 g, i.e. |S.sup.-1(U-O)|=1 g,
or |S.sup.-1 (U-O)|.sup.2=1 g.sup.2.
[0051] Such a rest state without interfering acceleration is
recognized in method step a. and checked in method step c.
[0052] In the present example, the assumed rest state is checked in
method step c. by implementation as a filter with the aid of a
so-called pseudomeasurement. For this purpose, the mean of the
acceleration is estimated, and the "measured value" is assumed to
be 1 g. The statement that a measured value of 1 g is assured as
the acceleration refers to the measurement equation, and thus, that
a "measured value" in the measurement equation does not have to
have been actually ascertained by a measurement. This is not a
measurement in the literal sense, since the value "1 g" is not
detected by a sensor. An innovation, i.e., a difference between a
measured "measured value" and an estimated "measured value," is
then monitored. In this example, a residuum (y-y') is selected as
an innovation, y being the actual measured value and y' being the
estimated measured value. Depending on the measurement equation
selected, y and y' may be scalars or vectors. In this case the
residuum is normalized to its standard deviation, which then must
correspond to a random variable having a normal distribution. A
test for normal distribution, specified by the mean value and the
variance, may then be applied to the random variable. If the
residuum is too large, an interfering acceleration is assumed, and
the measurement must be discarded with regard to a calibration.
[0053] In this case modeling is carried out using a parameter model
which is based on perturbations in the form of noise terms for
sensitivity and offset. Supplementation with a kinematic model also
takes a noise term into account for the acceleration, but this is
not addressed here. In the parameter model, a state change between
two successive measurements is described for sensitivity and
offset, using k as a running index: S.sub.k+1=S.sub.k+v.sub.S,k and
O.sub.k+1=O.sub.k+v.sub.O,k, with noise terms v.sub.S,k for
sensitivity and v.sub.O,k for offset. The noise terms may be used
to describe a variation over time, for example due to temperature
or aging, using a random walk model. According to the present
invention, residuum e of the pseudomeasurement is then e.sub.g,k=1
g-|S.sub.k.sup.-1 (u.sub.k+V.sub.k-O.sub.k)|, or, in metric units,
e.sub.g,k=9.81-|S.sub.k.sup.-1(u.sub.k+v.sub.k-O.sub.k)|, where v
stands for measurement noise of the sensor.
[0054] In flow chart 43, FIG. 4 shows a method for self-adjustment
of a triaxial acceleration sensor during operation in a different
specific embodiment from FIG. 1, in which identical method steps
have the same reference characters, and modified method steps have
apostrophes following the reference characters. The method steps
are as follows: [0055] a. ensuring the observability of the
sensitivity and the offset of the sensor, having the following
substeps: [0056] a1. recognizing a rest situation--if a rest
situation is not present, no calibration may be carried out and the
method is repeated according to arrow 44; [0057] a2. recognizing
whether new information is present--if new information is not
present, no calibration may be carried out, and the method is
repeated according to arrow 45; [0058] b. calibrating the sensor
with the aid of calibration values for sensitivity and offset;
[0059] c. checking the self-adjustment for an interfering
acceleration with the aid of a measurement equation and estimated
values for sensitivity and offset, having the following substeps:
[0060] c1. estimating sensitivity and/or offset and the variance
thereof; [0061] c2'. determining an innovation as the difference
between a measured value of the measurement equation and an
estimated value of the measurement equation--here as well, the
absolute value of an acceleration vector is estimated to be 1 g,
and a measured value of 1 g is assumed as the acceleration. A
normalized innovation is also used in this specific embodiment. The
residuum (y-y') is normalized to its variance Var(y-y'). The
innovation is then NIS=(y-y').sup.T Var'(y-y') (y-y'), with the
transposed residuum being (y-y').sup.T. This is followed by method
step [0062] c3'. testing the innovation for a chi square
distribution--this is coordinated with the use of a normalized
innovation in method step c2; [0063] c4'. recognizing the
interfering acceleration in the event of a deviation from the chi
square distribution; this is followed by method step [0064] d.
repeating the adjustment if an interfering acceleration is
recognized, as indicated by arrow 41'; [0065] e. accepting the
estimated values for sensitivity and offset as calibration values
if an interfering acceleration is not recognized. The method is
repeated during operation as indicated by arrow 42'.
* * * * *